基于经验模态分解的煤炭消费量组合预测

Combined prediction method of coal consumption based on empirical mode decomposition

为了明确煤炭消费量的发展变化规律和趋势,科学引导煤炭行业健康、有序发展。针对煤炭消费量的时间序列变化具有增长性、波动性和非平稳性的特点,采用经验模态分解法对1978-2014年煤炭消费量进行多层次分解,得到其发展变化的趋势量和波动量。利用BP神经网络的非线性映射能力,分别对趋势量和波动量进行预测,最终二者相加求和得到煤炭消费量的预测值。误差分析表明,基于经验模态分解的煤炭消费量组合预测模型,拟合值的平均误差为2.18%,预测值的平均误差为1.24%。该组合预测模型可以有效的提高煤炭消费量的预测精度,用该模型预测了2015-2020年煤炭消费量。预测结果表明,在未来几年煤炭消费量将保持低速增长趋势,到2020年将达到341718.2万t标准煤。

This paper aims to investigate the law behind how coal consumption tends to develop in an effort to guide the healthy and orderly development of coal industry. In response to the growth,volatility,and instability inherent in the time series of coal consumption,the research using empirical mode decomposition method consists of the multi-level decomposition of the time series of coal consumption from 1978 to 2014 to obtain its development trend quantity and fluctuation quantity; the prediction of the trend quantity and fluctuation quantity using the nonlinear mapping ability of BP neural network and ultimate achievement of the forecasting results of coal consumption by summing the two forecast results. The error analysis shows that combined prediction model of coal consumption based on empirical mode decomposition gives an average error of 2. 18% for fitted values and the average error of 1. 24% for predicted values. The model enables an effective improvement in the forecast accuracy of coal consumption and works better for predicting coal consumption between 2015 and 2020. The prediction show that the coal consumption tends to keep a moderate growth in the coming years,and is expected to reach 3. 417182 billion tce by 2020.